Figure AI is pivoting away from traditional hand-coded software to train its humanoid robots using neural networks that learn through vision and language, a fundamental shift that CEO Brett Adcock believes positions the company to achieve artificial general intelligence (AGI) through physical embodiment. The San Jose-based company, which operates a four-building campus with 250 to 300 employees, designs, manufactures, and tests every robot entirely in-house, moving through rapid hardware iterations while betting everything on a software-first philosophy.

What Makes Figure's Helix Neural Network Different?

Rather than writing hundreds of thousands of lines of traditional C++ code to control the robots' 40+ joints, Figure developed Helix, a proprietary Vision Language Action model that allows robots to reason through pixel space and translate camera inputs into physical movements. The system runs autonomously on the robot's internal graphics processing units (GPUs), meaning the robots do not require a Wi-Fi connection to function and are not teleoperated by humans.

The company trains its robots in simulation environments similar to video games, then transfers that programming directly to physical robots with remarkably high success rates. This sim-to-real approach dramatically accelerates development compared to traditional robotics, where engineers typically hand-code responses to specific scenarios. Adcock explained that this neural-network-first strategy represents a complete departure from how the company initially approached robot control, reflecting a broader industry shift toward AI-driven autonomy.

How Does Figure's "Never Fall" Safety System Work?

One of the most striking demonstrations during the campus tour was Project Vulcan, a fault-tolerant control system that allows Figure robots to survive critical hardware failures. If a robot loses power or communication to a joint, such as a knee, the system can lock that joint's velocity and allow the robot to hobble to safety or even continue working without toppling over. This capability addresses one of the biggest challenges in deploying humanoids in real-world environments: the ability to gracefully degrade when something goes wrong.

The controls team stress-tested this system extensively, pushing robots to their physical limits with burpees and deliberate shoves to identify failure modes before models are released to customers. This "trust but verify" mindset reflects Figure's commitment to reliability over speed, a notable shift in an industry often criticized for prioritizing flashy demos over practical durability.

The Evolution From Figure 01 to Figure 04

Figure has rapidly iterated through multiple robot generations, prioritizing speed of development over initial costs. The company's design studio houses every generation ever built, revealing the dramatic evolution of the platform:

• Figure 01: Built entirely with expensive CNC-machined parts to get hardware to the software team quickly within the first year, featuring a tendon-driven hand that mimicked human biology but proved impractical, forcing a pivot to a bulky "Frankenstein" wrist using repurposed motors.
• Figure 02: Moved the battery from an external backpack into the torso, tripled computing power, and featured an exoskeleton structure where the exterior skin handled load-bearing, but was retired due to manufacturing complexities and frequent reliability issues.
• Figure 03: Achieved a 90% cost reduction compared to Figure 02, featuring a slimmer profile, soft-wrapped foam exterior, and new tactile hands equipped with cameras, plus custom high-top sneaker-style feet designed with openings that push air through the legs for cooling during wireless charging.
• Figure 04: Currently in detailed design phase and expected to represent a massive leap forward, which Adcock describes as the company's upcoming "iPhone 1 moment" representing what they believe will be the perfect humanoid form factor.

"Figure 04 will be their 'iPhone 1 moment,'" stated Brett Adcock, Founder and CEO of Figure AI.

Brett Adcock, Founder and CEO, Figure AI

The progression from Figure 01 to Figure 03 demonstrates the company's willingness to abandon expensive approaches in favor of manufacturing efficiency. The 90% cost reduction in Figure 03 is particularly significant for a company planning to deploy thousands of robots into commercial and residential environments.

How Is Figure Manufacturing Robots at Scale?

The BotQ manufacturing facility, located on Figure's campus, handles the full assembly process for Figure 03 robots. The facility includes custom-built battery production lines and end-of-line burn-in bays where every robot is tested before deployment. This vertical integration means Figure controls quality at every stage, from component manufacturing to final testing, reducing reliance on external suppliers and allowing rapid iteration when design changes are needed.

The company's commitment to in-house manufacturing reflects a broader trend in the robotics industry toward controlling the entire supply chain, particularly as companies race to scale production. With 18,000 humanoid units shipped globally in 2025, roughly 90% from Chinese manufacturers, Western companies like Figure are betting that vertical integration and proprietary AI will be their competitive advantage.

What Are Figure's Real-World Deployment Plans?

Figure has already demonstrated its robots performing autonomous tasks in home environments, tidying living rooms using the Helix neural network without any human teleoperation. The company is also pursuing industrial deployments, including work on automotive assembly lines at BMW factories. These real-world tests are critical for validating whether the neural-network-first approach can handle the unpredictability of actual work environments, not just controlled lab settings.

The company offers robots on a leasing model, positioning them as 24/7 workforce solutions for customers who need continuous operation. This business model differs from traditional robotics sales and reflects the industry's shift toward robotics-as-a-service, where customers pay for robot hours rather than purchasing hardware outright.

How Does Figure Fit Into the Broader Humanoid Robotics Consolidation?

Figure operates in a rapidly consolidating market. In 2025, there were 12 mergers and acquisitions deals exceeding $50 million in the humanoid robotics space, compared to just 4 in 2024. The industry has attracted $8.6 billion in funding through mid-2026, already exceeding all of 2025's total funding. Major technology companies are making strategic bets: Meta invested $1 billion for a 49% stake in Aria Robotics in May 2026, while Amazon has invested in multiple humanoid startups including Fauna Robotics and Mentee Robotics.

Unitree Robotics, a Chinese competitor, became the first pure-play humanoid company to go public when it listed on Shanghai's STAR Market in July 2026 with a $6.2 billion valuation. However, the U.S. Federal Communications Commission's ban on foreign-produced humanoid robots, announced in late July 2026, adds significant regulatory headwinds for international expansion of Chinese manufacturers. This regulatory environment may benefit Western companies like Figure that manufacture domestically.

The projected market for humanoid robotics could reach $13 trillion by 2050 according to RBC Capital, creating enormous incentives for companies to establish market leadership now. Figure's focus on general-purpose robots that can handle diverse tasks, combined with its proprietary AI and manufacturing capabilities, positions it as a key player in this emerging market.

Steps to Understanding Figure's Technology Advantage

• Vision-Language-Action Models: Figure's Helix system processes camera inputs and translates them into physical movements, allowing robots to understand their environment and act autonomously without constant human control or pre-programmed responses.
• Simulation-to-Reality Transfer: The company trains robots in physics simulators before deploying them physically, dramatically reducing development time and allowing rapid iteration on control algorithms without damaging expensive hardware.
• Fault-Tolerant Design: Project Vulcan enables robots to survive hardware failures and continue operating, addressing a critical reliability concern that has limited humanoid deployment in unpredictable real-world environments.
• Vertical Manufacturing Integration: By controlling design, manufacturing, and testing in-house, Figure can iterate quickly and maintain quality standards while reducing dependence on external suppliers in a competitive market.

Figure's approach represents a fundamental rethinking of how to build humanoid robots. Rather than perfecting hardware first and then adding software, the company prioritizes AI development and uses hardware iteration to support software capabilities. This philosophy aligns with broader trends in artificial intelligence, where neural networks and machine learning have proven more effective than hand-coded rules for handling complex, unpredictable environments.

As the humanoid robotics industry consolidates and major technology companies make strategic investments, Figure's combination of proprietary AI, manufacturing capability, and real-world deployment experience positions it as a significant contender in what could become one of the largest technology markets of the next decade.